EEG motor imagery classification through a two-dimensional CNN-LSTM deep architecture and fuzzy decision-making

This study presents a robust deep learning framework for automatic motor imagery detection from raw EEG signals. Six band-power features were extracted using STFT, and dedicated 2D CNN-LSTM models were trained for each band. Their outputs were fused using a Choquet fuzzy integral to enhance decision reliability under noisy EEG conditions. Alpha- and sigma-band models achieved 88% and 87.1% accuracy, respectively. The fused architecture reached 90.4% on BCI IV-2a and 92.21% on BCI IV-1, outperforming existing methods in motor imagery classification.